Docstoc

Kalman Filtering and the

Document Sample
Kalman Filtering and the Powered By Docstoc
					                             AIAA-2004-6018
                         Kalman Filtering and the
                 Attitude Determination and Control Task
                       C1C Matthew J. Hale*, Dr. Paul Vergez‡ and Maarten J. Meerman§
                        USAFA, Department of Astronautics, USAF Academy CO 80840

                                               Dr. Yoshi Hashida†
                     Surrey Satellite Technology Ltd., University of Surrey, United Kingdom



            The attitude determination and control subsystem requirements of various payloads on
         the FalconSAT-3 satellite are stringent enough to demand some type of data processing in
         order to meet attitude determination requirements. This paper details one data filter, the
         Kalman filter, and more specifically, the 7-state and 6-state Kalman filters. Both filters
         prove to meet the attitude determination requirements successfully with little difference in
         achieved accuracy. However, the 6-state filter places much less computational demand upon
         the on-board computer. Based upon this data, the 6-state filter is a more logical choice for
         the FS3 attitude determination and control subsystem filter. However, there is still much
         analysis that needs to be completed before a final verdict might be reached. This paper
         examines both of these Kalman filters from a theoretical standpoint before examining
         practical implementation and concerns therein.


                                                 Nomenclature
Ax      = attitude matrix (transition matrix)
I         = inertia tensor
N         = disturbance torques
P         = covariance matrix
qn        = quaternion
TBYLO     =   attitude matrix (transition matrix)
x         =   propagated “x” vector
ˆ
x         =   corrected “x” vector
x         =   state vector
q n      =   differential quaternion
        = differential angular rate
   I
    By                                       
          = Euler body rates wrt. inertial,  x      y  z t
 Lo
  I
                                                    
          = Euler local orbital rates wrt. inertial, ox    oy oz t
o        = mean motion of the Earth,   0    o   0
                                                        t




*
  Cadet First Class, Department of Astronautics, PO Box 4008 USAF Academy CO 80841.
‡
  Professor, Department of Astronautics, USAF Academy CO 80840. AIAA Associate Fellow
§
  Schriever Professor, Dept. of Astronautics, USAF Academy CO 80840. AIAA senior Member
†
  Chief ADCS Engineer, Surrey Satellite Technology Ltd., Guildford Surrey GU2 7XH, United Kingdom.

                                               1
                          American Institute of Aeronautics and Astronautics
                                                 I. Introduction

I  N 1990 the Space Systems Research Center (SSRC) was founded to enable cadets at the United States Air Force
   Academy to fully experience the complexity of satellite design, construction, and operation. After several
   successful balloon-based upper atmosphere experiments, SSRC initiated the development of operational satellites
for Air Force research experiments. In 1997, SSRC‟s FalconGold satellite bore a GPS Signal Experiment into an
elliptical orbit above the GPS constellation. FalconGold successfully completed its mission of collecting data
showing that GPS signals can be received above the GPS constellation and be utilized for orbit determination.
FalconSAT-1 was launched in 2000 to assess the hazards of spacecraft operations in the wake of larger bodies.
However, an in-orbit power failure prevented the successful completion of this mission. FalconSAT-2 was designed
to gather data concerning the effect of upper-atmospheric plasma bubbles upon GPS signals. Originally designed for
a 2003 launch in conjunction with the Space Shuttle Hitchhiker program, the satellite is currently being re-
engineered for an alternate launch vehicle due to the probable cancellation of the Hitchhiker program.
    FalconSAT-3 (FS3), SSRC‟s current satellite design endeavor, is a 50 kg microsatellite being developed by
cadets and faculty at the Air Force Academy. FS3 will carry several experimental payloads to conduct DoD
research. One of these, the Micro Propulsion Attitude Control System (MPACS), is an attempt to demonstrate
pulsed plasma thruster (PPT) propulsion technology. The Flat Plasma Spectrometer (FLAPS) and the Plasma Local
Anomalies Noise Environment (PLANE) experiments are further attempts at characterizing the local plasma
environment. A shape memory composite gravity gradient boom and a shock ring vibration suppression system
comprise the final two experimental payloads of FS3. The satellite is scheduled to be launched in the Fall of 2006
for a projected design to end-of-life cost of approximately $2 million.

                                          II. FS3 ADCS Architecture
     The primary challenge for the FS3 Attitude Determination and Control System (ADCS) team is to develop the
software and techniques necessary to achieve the derived attitude control requirements. As it turns out, the FLAPS
experimental payload has the most stringent ADCS requirements, demanding attitude control to within five degrees
of the ram direction with respect to the local orbital frame, as well as attitude knowledge to within one degree. FS3
is the Air Force Academy‟s first attempt at achieving three axis attitude control.
     FS3 attitude sensors include one Billingsby Fluxgate magnetometer and 4 AeroAstro medium sun sensors. In
conjunction with on-board software (some of
which is detailed in this paper), these sensors
will provide sufficient attitude knowledge to
meet the derived requirements. Attitude actuators
include a gravity gradient boom for passive
attitude control, as well as three magnetorquers
for active attitude control. The boom is a 2.84
meters long (to the center of gravity) composite
thermal boom with a 7.8 kilogram tip mass,
creating a transverse moment of inertia of 67.4
kg-m-s2.
     As part of the ADCS task, a Kalman filter
will be implemented to process sensor attitude
date. Generally speaking, the Kalman filter is a
recursive optimization algorithm that generates
an estimate based upon potentially noisy
observation data. At the most basic level, the
Kalman filter is fundamentally an optimization
problem that can be applied across many
disciplines to predict the behavior of systems. In
astronautics, the Kalman filter is often                  Figure 1. EKF Mechanics, One Measurement
implemented to simplify and expedite the ADCS
task. For FS3, the filter must balance ADCS
sensor measurements with the expected attitude generated by an on-board orbit propagator. For several reasons, the
Kalman filter is implemented in this process rather than other filtering techniques. Due to on-board computer (OBC)

                                                2
                           American Institute of Aeronautics and Astronautics
limitations, data must be cyclically processed and discarded, rather than being stored for access during each
estimation iteration as with other
     filtering techniques. The Kalman filter is ideal for processing large amounts of data in this fashion. Attitude
determination presents another difficulty that is addressed by the Kalman filter. In the attitude determination task,
three independent reference parameters are needed to determine attitude. Each vector measurement provided by the
satellite sensors yields two reference parameters. Therefore, the typical requirement for three-axis attitude
determination is two vector measurements. Attitude determination with two measurements is overdetermined while
attitude determination with one parameter is underdetermined. The Kalman filter can solve either the
overdetermined or the underdetermined cases, allowing three-axis attitude determination with either both sun
sensors and magnetometers or solely magnetometers. The Kalman filter is able to solve the underdetermined case
because an on-board orbit propagation model is part of the filter. This filter utilizes this model to allow three-axis
attitude determination with only one vector measurement. In short, the robustness of the Kalman filter enables it to
be applied to the ADCS task.
     The estimation of the Kalman filter operates in two primary cycles, propagation and correction. During the
propagation cycle, the filter propagates the state of the system, using a system model to predict the state of the
system one time step in the future. The correction cycle inputs measurements of the system state and utilizes these
observations to correct for differences between the state propagated from the system model and the measured
satellite state. However, the correction cycle encounters particular difficulty due to the fact that some amount of
noise and imprecision is embodied in the measurements themselves. Therefore, the primary task of the Kalman filter
correction cycle is to balance the state propagated from the system model with the system state derived from
measurements utilizing optimization theory. This correction process yields a „corrected‟ estimate of the system state.
As the filter iterates, this corrected state estimate is utilized as the initial condition for the Kalman filter propagation
cycle.
     Figure 1 offers a pictorial representation of basic Kalman Filter dynamics. Notice that the covariance matrix is
propagated and corrected as well. This matrix contains information essential to the optimization process, and
therefore must be included in the propagation – correction process.
     However, FS3 attitude knowledge hardware includes both a magnetometer and a sun sensor. Hence, two
measurements are entered to the filter when the satellite is illuminated. This does not significantly affect the basic
filtering process, yet it merits discussion. Magnetometer measurements are assumed less accurate than the sun
sensor measurements. Therefore, magnetometer measurements are first entered into the correction cycle of the EKF,
resulting in a „corrected‟ state. Subsequently,
before returning to the propagation cycle, the sun
sensor measurements are entered into a second
iteration of the correction cycle, yielding the
further „corrected‟ state. The diagram below
offers a pictorial representation of the dynamics
of dual measurement input to a Kalman Filter.
     For ADCS, Kalman filtering involves
propagation of the satellite attitude and
covariance matrices using both Euler‟s moment
equations and a basic knowledge of the
disturbance torques acting upon the satellite.
Subsequent to this propagation, the Kalman filter
adjusts the propagated attitude and covariance
matrices based upon the measurement vector(s).
Because the attitude motion of FS3 will be
nonlinear, extended Kalman filters (EKF) will be
necessary to accommodate nonlinearities. This
paper will detail the various filtering schemes
explored for optimal application in the FS3
ADCS task.                                                   Figure 2. EKF Mechanics, Two Measurements

           III.    Seven State EKF
   The EKF most commonly implemented in the ADCS task is the 7-state EKF. This is due, in part, to the relative
ease of 7-state mathematics in comparison with other Kalman filters that have been utilized for spacecraft


                                                  3
                             American Institute of Aeronautics and Astronautics
operations. In addition, the 7-state filter has well-established reputation as an effective and reliable estimation
technique.

A. Theoretical Detail
     For the 7-state filter, the state vector defines not only the attitude of the satellite, but also the rates at which the
attitude is changing. The 7-state EKF state vector is comprised of the four-element quaternion attitude vector
combined with the three-element body rates vector, with respect to the inertial frame        . Symbolically, this state
                                                                                              I
                                                                                              By
vector can be represented as

                                           
                                      x  q1      q2    q3    q4  x       y  z t                                     (1)

   During the propagation cycle of the EKF, the quaternion and angular rate components of the state vector are
propagated separately. The quaternions are propagated forward in time utilizing the basic quaternion dynamic
equation, specifically

                                                             1
                                                        q
                                                              q                                                        (2)
                                                             2

                                                0             oz       oy     ox 
                                                             0        ox      oy 
                                     where                                         
                                                    oz
                                                                                                                           (3)
                                                 oy          ox       0       oz 
                                                                                   
                                                 ox
                                                              oy      oz     0 
                                                                                    

                                                   lo    TBYLO  o                                                     (4)


                                                Lo  ox oy oz t
                                                 I
                                                                                                                           (5)


                                                  o  0  o 0t                                                         (6)

The body rates are propagated forward in time utilizing Euler‟s Moment Equations, specifically

                                                    I  N    I
                                                                                                                               (7)

Simple numerical integration is utilized to propagate the quaternions and body rates one time step in the future.
    With these definitions in mind, the basic mathematical process for 7-state Kalman Filtering can be detailed
below. Note that several equations integral to the operation of a Kalman Filter are here overlooked in order to
simplify the presentation. Many others have presented the EKF in much more detail, and the omitted equations may
be found in these works.*




*
    Hashida, Yoshi. ADCS for Future UoSat Standard Platform: REVISION 2. SSTL Internal Technical Note.
     SSTL, Guildford, Surrey, United Kingdom: 2004.
    Plessis, Roger M. Poor Man‟s Explanation of Kalman Filtering or How I Stopped Worrying and Learned to Love
     Matrix Inversion. Rockwell International, CA: June 1997.

                                                  4
                             American Institute of Aeronautics and Astronautics
B. 7-State Kalman Filter Mathematical Process
       
x  q1 q2 q3 q4  x  y  z                              t


P - covariance matrix (7 x 7)                        K - Kalman gain matrix (7 x 3)
 - state transformation matrix (7 x 7)              R - measurement noise covariance matrix (3 x 3)
Q - process noise covariance matrix (7 x 7)          H - observation matrix (3 x 7)
F - mathematical convention (7 x 7)                  t - time
z - measurements of system state, either sun-sensor or magnetometer
z by - body referenced measurements, directly from onboard sensors
z lo - orbit referenced measurements, from orbit model prediction (IGRF)

A. Propagation Cycle                                                    B. Correction Cycle
 1. Covariance Propagation                                               1. Compute Observation Matrix
                 ˆ
Pk 1   k 1 Pk  tk 1  Qk 1                                               z by
                                                                         Hk 
                                                                               x
                    q        q 
                                  
                    q                                                        T
                              by                                           BYLO z lo
                                                                                                TBYLO         TBYLO          TBYLO
                                   t k 1  t k 
                                                                                                       z lo           z lo            z lo ...
 k 1  I 7 x 7                 
                                                                               q1              q 2           q3             q 4
                     by  by 
                    q       by                                               TBYLO         TBYLO         TBYLO 
                                                                                       z lo           z lo          z lo 
                                                                                    x            y            z      
                                                                                                                           
            I 7 x 7  Fk  t
 2. Propagate State                                                     2. Compute Kalman Gain Matrix
                              t k 1                                                          t
                                                                                                   
                                                                                   K k  Pk H k H k Pk H k  Rk
                                                                                                         t
                                                                                                                        1

             x k 1  x k 
                      ˆ         xdt
                                 
                                                                         3. Update State
                                                                                    x k 1  x k 1  K k 1 z by  TBYLO z lo 
                               tk
             specifically
                                                                                    ˆ
                                    t k 1                               4. Update Covariance
                                        q dt
                             1
             q k 1    qk 
                        ˆ
                             2                                              Pk 1  I 7 x7  K k 1 H k 1 Pk 1
                                                                            ˆ
                                       tk
                               t k 1

              k 1   k        I N    I dt
                                            1
                      ˆ
                                 tk


   This basic EKF process has introduced several variables that merit further explanation. The covariance matrix
P  essentially is a time-referenced estimate of the accuracy of both the system model and the measurements. The
correction cycle depends heavily upon these accuracies in order to determine how much to „trust‟ either the
propagated state or the entered measurements. The state transformation matrix   is an approximation of the
change that the state undergoes over the specified time interval. The process noise covariance matrix Q  is derived
from the expected error in the filtering process. The observation matrix H  is a measure of how dependent the
measurements are upon the state of the system. The measurement noise covariance matrix R  entails the expected
error in the states themselves, derived from the precision of the system model.
          It is important to note several key aspects to the filtering process. Notice that the covariance matrix is
propagated in addition to the actual state. In the cyclic pattern of the EKF, the filter utilizes only the covariance
matrix and state from the previous iteration, which means that a relatively small amount of stored data points will
enable filter operation. This greatly reduces the computational demand of the filter. This is especially useful because
the covariance matrix and the state vector contain all the information concerning the status of the system that is
necessary for accurate system modeling.
    Another aspect worthy of note requires a basic understanding of the onboard application of the EKF. Onboard
FS3, the EKF will input sensor measurements every five seconds. Since the correction cycle occurs only once for

                                                                 5
                                            American Institute of Aeronautics and Astronautics
each set of measurements, the propagator must propagate the system state five seconds in the future, the point at
which the next set of measurements will be entered. However, the propagation cycle entails the numerical
integration of the system state over this time
period, which is the heaviest computational
demand placed by the system on the onboard
computer. In order to reduce this computational
demand, a simplified numerical integration
technique may be introduced, as long as the
resultant error is within acceptable bounds.
Integration time step may also be decreased in
order to reduce the computational demand,
contingent upon acceptable error bounds.
    As noted previously, if sun sensor
measurements are entered to the filter in addition
to the magnetometer measurements, the
correction cycle must be repeated twice. The first
iteration will update the state derived from the
most recent propagation. The second iteration,
utilizing sun sensor measurements, will update
the „corrected‟ state derived from the first
iteration of the correction cycle. Both the single  Figure 3. EKF Cycle
and double correction EKF cycles are
represented pictorially below.

                                              IV.     Six State EKF
    The 7-state EKF presents an optimal means of attitude control for small satellites. However, the 7-state EKF
necessarily involves many 7 x 7 matrices, placing a fairly heavy computational demand upon the onboard computer
of the typical microsatellite. Hence, any means of significantly reducing the computational intensity of the EKF
while not sacrificing operability is highly desirable.
    Recentely, Surrey Satellite Technology Ltd. developed a 6-state EKF that attempts to realize this goal.
Developed for BilSAT, a Turkish satellite member of the international cooperative effort of the Disaster Monitoring
Constellation, initial development of this 6-state EKF began in 2003. Currently, BilSAT is the only satellite known
to the author to have flown this ADCS software.*

A. Theoretical Detail
    In order to understand the theory underlying the development of the 6-state EKF, it is first necessary to touch
upon some fundamental quaternion definitions. In quaternion mathematics, the  operator is typically defined such
that

                                           p4          p3      p2      p1   q1 
                                           p          p4       p1      p 2  q 2 
                                    pq   3                                                                 (8)
                                           p2          p1      p4      p3   q3 
                                                                             
                                            p1        p2     p3      p 4  q 4 

where both p and q are quaternion vectors. This definition is intended to allow

                                              A p Aq   A p  q                                                 (9)




*
    Hashida, Yoshi. BilSat Attitude Estimator Mathematical Specification: REVISION 2. SSTL Internal Technical
     Note. SSTL, Guildford, Surrey, United Kingdom; 2004.

                                                 6
                            American Institute of Aeronautics and Astronautics
where A  is an attitude (transformation) matrix. This relationship proves pivotal to the derivation of many of the
relationships utilized in the 6-state EKF. Another convention utilized extensively in the 6-state EKF is q and  ,
the differential error notation, which represent small deviations in the quaternions and body rates, respectively. The
formal definition of these parameters is

                                                   qk 1  q  qk
                                                                                                                           (10)
                                                    k 1   k  

The        relationship is somewhat intuitive and simply understood, whereas the q  q relationship may not
be quite so intuitive. However, the q  q relationship as expressed above is very similar to the simple   
relationship, just made more complex by quaternion mathematics.
    Strictly speaking, true quaternions are always normal, that is,

                                                q12  q 2  q3  q 4  1
                                                        2    2     2
                                                                                                                           (11)

In order for the q  q relationship to hold and the  operator apply, q must be assumed to be a true quaternion.
(As the filter converges, this assumption becomes essentially valid.). Therefore, it follows that the differential
quaternions apply to Eq. 11. Namely, that

                                              q12  q 2  q3  q 4  1
                                                        2     2      2
                                                                                                                     (12)

This normalization allows any three q terms to necessarily define the fourth,


                                            q4   1  q12  q2  q3
                                                                 2     2
                                                                                                                     (13)


Note that this assumption defines q 4 as positive. This is acceptable since q 4 is a differential quaternion
specifying the error associated with the estimated quaternions, and the magnitude of the differential quaternion is the
essential information derived from the differential quaternions. It is important to note that q 4 is a differential
quaternion; therefore, it does not directly specify the attitude of a spacecraft. This assumption (that the differential
error quaternions meet the normalization criteria) is a significant assumption allowing the development of the 6-state
EKF.
    In the 6-state EKF, the auxiliary state vector is introduced, defined as the differential state, specifically

                                        
                                    y  q1 q2 q3 x y                  z t                                  (14)

Note that due to the assumption made in Eq. 13, the auxiliary state vector is comprised of only 6 elements. Based
upon the normalization assumption described above, and if higher order terms are ignored (linearization), it can be
demonstrated that q1  q 2  q3  0 . This relationship, in conjunction with the normalization assumptions, can
be applied with care to facilitate the simplification of a 7-state filter to a 6-state filter.
     Application of the definitions in Eq. 10 to the 7-state EKF equations allows the development of the following
filtering process. Note that the cross matrix definition is utilized extensively throughout this process. The cross
matrix is defined specifically as:




                                                 7
                            American Institute of Aeronautics and Astronautics
                                                                      0         az    ay 
                                                                                           
                               a  b  C a b             C a    a z       0      ax                             (15)
                                                                      a y      ax     0 
                                                                                           

   Two distinct state vectors are utilized in the 6-State EKF process. The true state is comprised of exactly the same
elements as in the 7-State EKF process, specifically

                                      x  q1    q2    q3     q4  x           y  z t                                 (16)

The 6-State filter also introduces the differential or auxiliary state, which is

                                         
                                     y  q1 q2 q3 x                 y z t
                                                                                                                   (17)

Note that the auxiliary state has six elements rather than seven. This plays a critical role in reducing the
computational demand of the 6-State EKF.

B. 6-State Kalman Filter Mathematical Process
     
x  q1      q2    q3    q4  x       y  z t                  
                                                          y  q1 q2 q3 x                   y z t
q -   differential quaternions                          -    differential body rates
P - covariance matrix (6 x 6)                         K - Kalman gain matrix (6 x 3)
 - state transformation matrix (6 x 6)               F - mathematical convention (6 x 6)
Q - process noise covariance matrix (6 x 6)           H - observation matrix (3 x 6)
R - measurement noise covariance matrix (3 x 3) t - time
z - measurements of system state, either sun-sensor or magnetometer
z by - body referenced measurements, directly from onboard sensors
z lo - orbit referenced measurements, from orbit model prediction (IGRF)




                                                  8
                             American Institute of Aeronautics and Astronautics
A. Propagation Cycle
 1. Covariance Propagation                                                      B. Correction Cycle
                           ˆ
           Pk 1   k 1 Pk  tk 1  Qk 1                                     1. Compute Observation Matrix
                                                                                       zby
                      q   q  
                                                                              Hk 
                      q                                                      y
 k 1    I 6 x6                        t  t 
                          
                                       k 1 k                                        T              TBYLO            TBYLO
                                                                                       BYLO zlo                zlo               zlo ...
                      q    
                                                                                      q1          q2            q3 
          I 6 x 6  Fk  t                                                             TBYLO           TBYLO            TBYLO 
                                                                                                                                   zlo 
                                                                                                            y 
                                                                                                 zlo                z lo
                                                                                       x                             z    
              C  
                                                          1
                  ˆ                                          I 3 x3
Fk                                                       2                
                                                                                      2  C TBYLO zlo  0 3 x 3 
              
      I 1 N  N
               gg     mt                         I C  I  C I  h 
                                                    1
                                                         ˆ          ˆ       
                                                                                 2. Compute Kalman Gain Matrix

N gg 
           6
               C u I  C Iu C u 
                                                                                                     t
                                                                                                           
                                                                                          K k  Pk H k H k Pk H k  Rk
                                                                                                                t
                                                                                                                                  1


            r3                                                                   3. Update Auxiliary State
N mt     2C m C Bby                                                                 y k 1  y k 1  K k 1 z by  TBYLO z lo 
                                                                                          ˆ
u  Aq 0 0 1
                                t                                               y k 1  K k 1 z by  TBYLO z lo 
                                                                                ˆ
                                                                                 4. Update Covariance
 2. Propagate State
                                    t k 1
                                                                                          Pk 1  I 6 x6  K k 1 H k 1 Pk 1
                                                                                          ˆ

                                      xdt
                                                                                 5. True State Update
             x k 1  x k 
                      ˆ                
                                     tk                                            q4  1  q12  q2  q3
                                                                                                       2     2

             specifically
                          t k 1
                                                                                q  q  q
                                                                                ˆ

                              q dt
                    1                                                               
                                                                                ˆ
    q k 1    qk 
               ˆ
                    2        tk                                                    x  q1 q 2 q3 q 4  x  y  z 
                                                                                   ˆ ˆ ˆ         ˆ   ˆ     ˆ   ˆ ˆ
                       t k 1

    k 1   k         I N    I dt
                                    1
            ˆ
                        tk


    It is important to note several aspects of the
6-state EKF process at this point. In addition,
key differences from the 7-state EKF will be
highlighted. Notice that both covariance and the
true state are predicted in the propagation cycle.
However, all matrices involved in the covariance
propagation are 6 x 6, which significantly
reduces the computational load upon the filter.
    Although this correction cycle may appear
very similar to that of the 7-state EKF, they are
markedly different. In the 7-state EKF, the state
vector is both propagated and directly updated.
That is to say, after propagation of the state
vector, the correction cycle acts directly upon the
state vector. The 6-state EKF process is slightly
more complex. The propagation cycle of the 6-
state EKF inputs the true state and propagates                                        Figure 4. 6-State EKF Cycle


                                                                  9
                                             American Institute of Aeronautics and Astronautics
this vector forward in time. The auxiliary state vector itself is not propagated. In the 6-state correction cycle, the
auxiliary (differential) state vector is calculated and then utilized to update the true state. Therefore, these
calculations primarily involve the auxiliary state. In this cycle, the true state is only utilized to calculate the attitude
matrix and in the last step to update the true state. This dependence upon the auxiliary state rather than the true state
plays a significant role in reducing the computational demand of the 6-state EKF. The 6-state EKF process may be
represented pictorially as below.
    The 6-state EKF, while mathematically more complex, reduces the covariance matrix P  , state transition
matrix   , F matrix, and the process noise covariance matrix Q  from 7 x 7 matrices to 6 x 6 matrices. In effect,
this eliminates 13 elements for each matrix. In           Table 1. EKF Element Comparison
addition, the observation matrix H  is reduced
from a 3 x 7 to a 3 x 6, and the Kalman gain matrix               7-                      6-
K    is reduced from a 7 x 3 to a 6 x 3. This                   State                   State
reduction in size is the great benefit of the 6-state                      Elements                 Elements
EKF. By assuming normalization and linearization          x       (7x1)         7          (7x1)          7
of the quaternion differential error, a much less         y          ~          ~          (6x1)          6
computationally intense EKF is possible. Note that        P       (7x7)         49         (6x6)         36
computational demand and processing speed have            Φ       (7x7)         49         (6x6)         36
not yet been empirically determined for either
filter. However, the table below demonstrates the
                                                          F       (7x7)         49         (6x6)         36
sheer magnitude of the reduction in the number of         Q       (7x7)         49         (6x6)         36
matrix elements, which should correlate with the          H       (3x7)         21         (3x6)         18
actual computational speed.                               K       (7x3)         21         (6x3)         18
                                                                      Total         245          Total         193
           V. Filter Implementation
     In order to examine its operability and performance, both the 6-state and 7-state EKF were programmed in the
computer language C. The data presented hereafter is primarily a summary of conclusions reached based upon a
series of analyses. It is intended to demonstrate the operability of the implemented EKFs as well as to provide a
basis for comparison of these EKFs with one another. In addition, issues that are important for practical
implementation yet unsuitable for the EKF theory discussion will be presented here.
     In order to test the EKF before launch, a simulation of the operating environment of the satellite must first be
developed. This simulation is required in order to generate realistic measurements for the EKF to utilize as input.
The simulation utilized to generate the input measurements for all analyses detailed herein is a third generation
flight-tested orbit propagator. It has been used on various satellites engineered by Surrey Satellite Technology Ltd.
This orbit propagator generates simulated magnetometer and sun sensor inputs in two different regards. First, the
orbit propagator generates these magnetometer and sun sensor inputs in the local orbital (with respect to inertial)
frame. These filter inputs are what is considered the orbit propagator „predicted‟ values. This input from the
simulation will be necessary during actual satellite operations. The second set of input data generated by the orbit
propagator is a set of „measured‟ values. These values simulate measurement input from the magnetometer and sun
sensor, and so this set of input is in the body (with respect to inertial) frame. Since this second set of inputs, the
„measurement‟ inputs, is only simulating values that will come from onboard sensors during the actual operation of
the satellite, this data will not be generated by the simulator during actual satellite operations.
     In addition, a standard of comparison is necessary in order to determine the veracity of the EKF results. Another
well-tested program developed by Surrey Satellite Technology Ltd. was utilized for this purpose. This simulation is
an attitude propagator, which models the actual satellite dynamics and outputs Euler angles and Euler rates over a
specified period of time. This simulation simply uses Euler‟s Moment Equations and the quaternion dynamic
equation to propagate expected angles and rates. Due to the construction of this model, it is quite simple to simplify
the model to discount all disturbance torques etc. or to precisely model gravity gradient (including boom
deployment), magnetic (including commanded magnetic moment), drag, reaction wheel effects, and other
disturbance torques.
     It is appropriate to mention several practical considerations of both EKFs implementation at this point. FS3 will
be operating in a 35˚ inclined circular LEO at an altitude of 560 km. The spacecraft is estimated to have a mass of
47.2 kg and essentially be configured as a cube 0.46 m on a side with a 2.84 m (to the center of gravity) gravity
gradient boom and 7.8 kg tip mass. The satellite will take attitude measurements every five seconds and requires 1˚
attitude knowledge while only 5˚ attitude control accuracy is required. The microsatellite is optimally nadir-pointing

                                                 10
                             American Institute of Aeronautics and Astronautics
and velocity vector tracking. These requirements specify several operating constants for the EKFs, specifically the
mean motion  o  , the measurement input frequency, and the radius of the orbit. These and a summary of other
operating constants are presented here.

                                          r  6938.137 km

                                          o  0.0010924576567 rad sec
                                          Tmsmt  5 sec

                                             67.4   0     0 
                                              0
                                          I      67.45 0  kg  m 2                                                        (18)
                                                             
                                              0
                                                    0   1.31
                                                             
                                          ui  0 0 1
                                                              t

                                                                      3
                                            3.986005e 5 km
                                                                          sec2
                                          tint  1 sec


Note that the assumed integration time step       t int        is 1 second unless otherwise specified. In addition, the
definition of the body frame for FS3 defines the unit zenith vector ui  as shown. The inertia tensor is boom-
deployed. These operating constants apply to all analyses unless otherwise noted.
    Several other practical considerations merit mention. The integration in the state propagation is accomplished by
numerical integration. This cyclic process is computationally intensive. Therefore, to reduce the computational
demand without sacrificing significant accuracy, an Adam‟s 2 nd order numerical integrator will be implemented
rather than a more complex but more accurate integrator. Reference the integration analysis completed by the author
for further discussion of the numerical integrator. Another practical consideration concerns the nature of the
computer code. Since the code will be implemented on a satellite, the code must be extremely robust. Most
importantly, the ADCS computer code must not crash the satellites‟ onboard computer at all expenses. Therefore,
error checks are conducted many times throughout each cycle of the EKF. If an error is detected, the ADCS
algorithm is immediately exited and a message relayed to satellite operators to allow human operators to deal with
the problem rather than the onboard computer crashing as a result of the error. Finally, the nature of the assumptions
underlying the development of the EKF permits a small amount of error to creep into the quaternion calculations. In
order to negate this error, the quaternions must be normalized after every instance where they are calculated. This
includes both the quaternion calculation in the state propagation step and in the state update step.
    Finally, several other initialization parameters deserve an explanation. The covariance matrix P  embodies an
approximate error associated with attitude estimates. The first three (6-state)/four (7-state) diagonal elements of the
covariance matrix represent the estimated error of the quaternions, while the next three diagonal elements give the
estimated error of the angular rates. It is important to note that the covariance matrix is merely an initialization; over
time, the matrix changes as the EKF converges. The process noise covariance matrix Q  is another initialization
parameter that deserves mention. The process noise covariance matrix contains information relating an estimate for
the error associated with the system equations. From a strict mathematical standpoint, the process noise covariance
matrix also changes with time. However, these changes can be ignored due to their small magnitude. The first three
(6-state)/four (7-state) diagonal elements of the process noise covariance matrix represent the error associated with
the quaternion dynamic equation, while the next three diagonal elements are the estimated error associated with
Euler‟s moment equations. Note that to rigorously implement this matrix, these error estimates are the subject of
several mathematical operations. The purpose of these operations is beyond the scope of this paper, but for
                                                                                          4
reference, the process noise covariance matrix error estimates for all cases is 1e for both quaternions and rates.
Lastly, the measurement noise covariance matrix R  contains the error expected to be associated with


                                                11
                            American Institute of Aeronautics and Astronautics
measurements. Note that the measurement noise covariance will change as the sun sensor is switched on and off.
However, the measurement noise covariance is otherwise constant over time.

                                                  VI.      Extended Kalman Filter Analysis

A. 7-State Analysis
   The following initialization parameters were utilized for 7-state analysis.
                      0.12     0      0       0              0                                                    0                        0        
                                                                                                                                                      
                       0      0.12
                                         0       0              0                                                    0                        0        
                       0         0    0.1 2
                                                 0              0                                                    0                         0       
                                                                                                                                                      
                       0         0      0     0.1 2
                                                                0                                                    0                         0       
                                                        o  
                                                                       2
                                                                                                                                                       
             P7 X 7   0         0      0       0       0.1                                                      0                         0       
                                                              180o                                                                                  
                                                                          o                                             
                                                                                                                               2
                                                                                                                                                       
                       0         0      0       0              0          0.1                                                              0       
                                                                                                                   180o                             
                                                                                                                                       o   
                                                                                                                                                     2

                       0         0      0       0              0                                                    0                  0.1        
                                                                                                                                             180o  


                           1e  4  T 3                                                                                                          
                                              0             0                          0                            0             0         0     
                            12  I x
                                      2

                                            4
                                                                                                                                                   
                                         1e  T     3
                                                                                                                                                   
                                  0                         0                          0                            0           0           0 
                                            12  I y2
                                                         4
                                                                                                                                                   
                                                       1e  T 3                                                                                  
                                  0           0                                        0                            0           0           0 
                                                         12  I z2                                                                                
                                                                      2 1e  4 2 1e  4 2 1e  4                                                
                                                                       q1        q2  2  q3  2   T 3
                 Q7 X 7                                                   I x2        Iy       Iz 
                                                                                                                                                   
                                  0           0             0                                                     0           0           0 
                                                                                         12                                                       
                                                                                                              1e  4  T                         
                                  0           0             0                          0                                        0           0 
                                                                                                                   Ix2
                                                                                                                                                   
                                                                                                                          1e  4  T             
                                  0           0             0                          0                            0                       0 
                                                                                                                                                  
                                                                                                                                  2
                                                                                                                                Iy
                                                                                                                                      1e  4  T        (19)
                                  0           0             0                          0                            0           0                 
                           
                                                                                                                                           I z2  
                           0.0252       0           0 
                                                          
                 RMT       0         0.0252        0 
                            0
                                          0       0.025 
                                                          2




The Kalman filter was run for a single twenty-four hour period, with the initial conditions detailed previously.
During this period, the sun sensor was switched off. In other words, all of these graphs present the case where
magnetorquers are the sole attitude determination sensors. In addition, all measurements are being normalized to
negate possible effects of the vector magnitude. Finally, gravity gradient disturbance torque was the only
disturbance modeled.
    The angular estimation error is presented graphically below. Note the distinctive „pulsating‟ effect of the 7 State
angular estimation error plot. These pulses arise from the magnetic field as the satellite orbits above the Earth. In
fact, if the pulses are tallied, there are fifteen, which corresponds to the number of times the satellite orbits the earth
in a single day. (At an altitude of 560 km, there are 15.02 orbits in solar day and 14.98 orbits per sidereal day).
These pulsations are therefore expected, due to the variation in the magnetic field around the Earth and its
interaction with the spacecraft dipole. In addition, the 7-state rate estimation error is presented here. Tabulated error
magnitudes are presented below the graphs.



                                                        12
                                    American Institute of Aeronautics and Astronautics
                                                                                                  F alco n S at3 : A D C S L o g F ile                                                                                                                                                                                       F alco n S at3 : A D C S L o g F ile
                                                                                                     7 S ta te A n g u la r E s tim a tio n E rro r                                                                                                                                                                               7 S ta te R a te s E s tim a tio n E rro r


                          R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                  P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                              W x E s t im a t io n E r r o r                                                                           W y E s t im a t io n E r r o r

                                                                                                                                                                                                                                                   0.06
    3

                                                                                                                                                                                                                                                   0.05

    2
                                                                                                                                                                                                                                                   0.04


    1                                                                                                                                                                                                                                              0.03


                                                                                                                                                                                                                                                   0.02
    0

                                                                                                                                                                                                                                                   0.01

   -1
                                                                                                                                                                                                                                                   0.00


   -2                                                                                                                                                                                                                                             -0.01


                                                                                                                                                                                                                                                  -0.02
   -3

                                                                                                                                                                                                                                                  -0.03

               19 T hu                      3A M                                   6A M                              9A M                         12P M                          3P M                                   6P M          9P M                       19 T hu        3A M                                  6A M                        9A M                          12P M      3P M                            6P M   9P M
A ug 2004                                                                                                                   T im e                                                                                                            A ug 2004                                                                                                   T im e



                        Figure 5. 7-State Angular Est. Error                                                                                                                                                                                                               Figure 7. 7-State Rates Est. Error
                                                                                                  F alco n S at3 : A D C S L o g F ile                                                                                                                                                                                       F alco n S at3 : A D C S L o g F ile
                                                                                                7 S ta te A n g u la r E s tim a tio n E rro r (D e ta il)                                                                                                                                                                   7 S ta te R a te s E s tim a tio n E rro r (D e ta il)


                            R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                  W x E s t im a t io n E r r o r                                                                        W y E s t im a t io n E r r o r
     0.4

                                                                                                                                                                                                                                                   0.00075
     0.3

                                                                                                                                                                                                                                                   0.00050
     0.2

                                                                                                                                                                                                                                                   0.00025
     0.1

                                                                                                                                                                                                                                                   0.00000
     0.0

                                                                                                                                                                                                                                                  -0.00025
    -0.1

                                                                                                                                                                                                                                                  -0.00050
    -0.2

                                                                                                                                                                                                                                                  -0.00075
    -0.3



              19 T hu                     3A M                                   6A M                               9A M                         12P M                          3P M                                   6P M           9P M                    19 T hu         3A M                                  6A M                        9A M                           12P M      3P M                             6P M   9P M
A ug 2004                                                                                                                     T im e                                                                                                          A ug 2004                                                                                                     T im e



            Figure 6. 7-State Angular Est. Error (Detail)                                                                                                                                                                                                    Figure 8. 7-State Rates Est. Error (Detail)

For the graphs presented above, notice that                                                                                                                                                                                        Table 2. 7-Sate Error Compilation
the filter takes on the order of 1-1.5 hours to
                                                                                                                                                                                                                                                        Avg d         STD d                                                                                                                        RMS d
converge. This represents the time it takes
from switching the filter on to the time at                                                                                                                                                                                        Roll (deg):         1.28E-03      6.99E-02                                                                                                                     6.99E-02
which the filter provides an accurate attitude                                                                                                                                                                                     Pitch (deg):       -7.25E-03      3.69E-02                                                                                                                     3.76E-02
estimate. The tabulated values will primarily                                                                                                                                                                                      Yaw (deg):         -1.62E-02      1.38E-01                                                                                                                     1.39E-01
be utilized for comparative purposes in                                                                                                                                                                                            Wx (deg/s):        -2.04E-05      1.52E-04                                                                                                                     1.53E-04
relation to the 6-state results. Note, however,                                                                                                                                                                                    Wy (deg/s):        -4.61E-05      1.35E-04                                                                                                                     1.42E-04
that the magnitude of the RMS error results                                                                                                                                                                                        Wz (deg/s):        -1.83E-06      1.74E-03                                                                                                                     1.74E-03
indicates that the angular error is
approximately 0.16 and the rate error is
about 0.0018, both in degrees. These results                                                                                                                                                                                                                     Mag of Err.                                                      Mag of Err.                                            Mag of Err.
fit within the bounds specified by the                                                                                                                                                                                             Angles (deg):                 -7.40E-03                                                         1.59E-01                                               1.60E-01
derived attitude requirements.                                                                                                                                                                                                     Rates (deg/s):                -2.28E-05                                                         1.75E-03                                               1.76E-03




                                                                                                                                            13
                                                                                                                        American Institute of Aeronautics and Astronautics
B. 6-State Analysis
    The following initial parameters were utilized for the 6-state EKF.

                                                                                                            0.12         0           0              0                      0                   0         
                                                                                                                                                                                                           
                                                                                                             0           0.12
                                                                                                                                        0              0                      0                   0         
                                                                                                             0             0       0.12             0                      0                   0         
                                                                                                                                                                                                           
                                                                                                                                                o  
                                                                                                                                                                 2
                                                                                                             0                                 0.1                                                       
                                                                                                                                                             o 
                                                                                                                            0           0                                     0                   0
                                                                                                   P6 X 6                                            180                                                
                                                                                                                                                                                                           
                                                                                                                                                                      o  
                                                                                                                                                                                      2
                                                                                                             0             0           0              0              0.1                      0         
                                                                                                                                                                             180o                        
                                                                                                                                                                                                          2
                                                                                                             0                                                                            o   
                                                                                                                            0           0              0                      0            0.1         
                                                                                                            
                                                                                                                                                                                                180o    
                                                                                                             1e  4  T 3                                                                                 
                                                                                                                                      0               0                 0            0              0      
                                                                                                              12  I x
                                                                                                                         2
                                                                                                                                                                                                            
                                                                                                                               1e  4  T 3                                                               
                                                                                                                    0                                 0                 0            0              0      
                                                                                                                                  12  I y2
                                                                                                                                                   4
                                                                                                                                                                                                            
                                                                                                                                                1e  T     3
                                                                                                                                                                                                            
                                                                                                                    0                 0                                 0            0              0      
                                                                                                                                                   12  I z2
                                                                                                   Q6 X 6                                                                                                 
                                                                                                                                                                  1e  4  T                              
                                                                                                                     0                 0               0                              0              0
                                                                                                                                                                       I x2                                
                                                                                                                                                                                  4                       
                                                                                                                                                                                1e  T                    
                                                                                                                     0                 0               0                 0                           0
                                                                                                                                                                                    Iy 2
                                                                                                                                                                                                            
                                                                                                                                                                                                           
                                                                                                                                                                                              1e  4  T 
                                                                                                                     0                 0               0                 0            0
                                                                                                             
                                                                                                                                                                                                   Iz2
                                                                                                                                                                                                            
                                                                                                                                                                                                            
                                                                                                            0.025    2
                                                                                                                                 0              0                                                                                                                                                                                                                                                                                                                    (20)
                                                                                                                                                     
                                                                                                   RMT   0                 0.0252           0 
                                                                                                             0
                                                                                                                                0          0.025 2




The 6-state angular estimation error is presented below. Notice that the 6-state angular estimation error plot does not
demonstrate periodic variation as the 7-state angular estimation error plot did. This is somewhat unexpected, as it
indicates that the 6-state EKF is able to produce slightly more accurate angular attitude estimate than the 7-state
EKF. Further testing is required to test the veracity of this claim. The 6-state rate estimation error plots are presented
below as well. Tabulated errors are presented below the graphs.
                                                                                              F alco n S at3 : A D C S L o g F ile                                                                                                                                                                                           F alco n S at3 : A D C S L o g F ile
                                                                                                6 S ta te A n g u la r E s tim a tio n E rro r                                                                                                                                                                             6 S ta te A n g u la r E s tim a tio n E rro r (D e ta il)

                          R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                      P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                   R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )
     3
                                                                                                                                                                                                                                   0.25


     2                                                                                                                                                                                                                             0.20

                                                                                                                                                                                                                                   0.15
     1
                                                                                                                                                                                                                                   0.10

                                                                                                                                                                                                                                   0.05
     0

                                                                                                                                                                                                                                   0.00

    -1
                                                                                                                                                                                                                                   -0.05

                                                                                                                                                                                                                                   -0.10
    -2
                                                                                                                                                                                                                                   -0.15

    -3                                                                                                                                                                                                                             -0.20

                                                                                                                                                                                                                                   -0.25
                 3P M           6P M                                    9P M                        19 T hu                         3A M                 6A M                                   9A M                   12P M
 18 W ed A ug 2004                                                                                                    T im e                                                                                                                3P M      6P M                                  9P M                                   19 T hu                         3A M                      6A M                                   9A M                      12P M
                                                                                                                                                                                                                               18 W ed A ug 2004                                                                                                          T im e


                        Figure 9. 6-State Angular Est. Error                                                                                                                                                                                       Figure 11.                                                               6-State Rates Est. Error




                                                                                                                                              14
                                                                                                                          American Institute of Aeronautics and Astronautics
                                                                     F alco n S at3 : A D C S L o g F ile                                                                                                                                 F alco n S at3 : A D C S L o g F ile
                                                                        6 S ta te R a te s E s tim a tio n E rro r                                                                                                                        6 S ta te R a te s E s tim a tio n E rro r (D e ta il)


                                  W x E s t im a t io n E r r o r                                                           W y E s t im a t io n E r r o r                                             W x E s t im a t io n E r r o r                                                                   W y E s t im a t io n E r r o r

    0.06

                                                                                                                                                                               0.00075
    0.05

    0.04                                                                                                                                                                       0.00050

    0.03
                                                                                                                                                                               0.00025
    0.02

                                                                                                                                                                               0.00000
    0.01

    0.00                                                                                                                                                                      -0.00025

    -0.01
                                                                                                                                                                              -0.00050
    -0.02

                                                                                                                                                                              -0.00075
    -0.03



                    3P M   6P M                               9P M           19 T hu                        3A M     6A M                               9A M   12P M                      3P M   6P M                        9P M                  19 T hu                         3A M            6A M                              9A M   12P M
18 W ed A ug 2004                                                                              T im e                                                                     18 W ed A ug 2004                                                                              T im e



    Figure 10.                                    6-State Angular Est. Err. (Detail)                                                                                               Figure 12.                                6-State Rates Est. Err. (Detail)

            Notice that the magnitude of the RMS error is within the derived attitude knowledge requirements demanded by
                                                                      
the FLAPS payload  1 . In addition, note that the magnitude of the RMS error of the 6-state EKF and the that of
                                                                                  

the 7-state EKF are almost exactly identical.

C. Analysis To Do
    Much work remains to be accomplished from the filtering standpoint of FS3.
              Program a constant gain filter for analysis and comparison
              Perform optimization analyses
                       o Initial Covariance Matrix (P)
                       o Observation Noise Covariance Matrix (R)
                       o Process Noise Covariance Matrix (Q)
                       o Boundaries for effects of measurement noise
              Minimize code size, memory allocation, computational demand,
              Maximize computational speed, redundancy, robustness
Once all of these tasks are accomplished, the selected filter will be tested on data derived from PICOSat. PICOSat is
a United States Air Force satellite that is solely controlled by Air Force Academy cadets trained to do so. Data from
PICOSat will be utilized due to the
similarity between PICOSat and FS3.                 Table 3. 6-State Error Compilation
PICOSat is a nadir-pointing, velocity vector
                                                                          AVG d           STD d         RMS d
tracking microsatellite with similar mass and
inertia tensor values. In addition, PICOSat         Roll (deg):        6.45E-03        3.07E-02       3.13E-02
utilizes a gravity gradient boom and                Pitch (deg):       -1.62E-03       4.70E-02       4.71E-02
magnetorquers for control, with sun sensors         Yaw (deg):         -3.24E-02       1.43E-01       1.47E-01
and magnetometers for attitude sensors.             Wx (deg/s):        1.04E-05        1.53E-04       1.53E-04
From the ADCS standpoint, these qualities           Wy (deg/s):        -5.45E-05       1.49E-04       1.58E-04
make PICOSat very closely related to FS3.           Wz (deg/s):        3.68E-05        1.94E-03       1.94E-03
Therefore, telemetry will be downloaded
from the satellite and then entered into both
EKFs to determine if their output accurately                           Mag of Err. Mag of Err. Mag of Err.
predicts the attitude dynamics of PICOSat.          Angles (deg): -9.20E-03            1.54E-01       1.57E-01
This test will lend credence to the validity of     Rates (deg/s): -2.43E-06           1.95E-03       1.96E-03
the software.

                                                                                                                                                               VII. Conclusions
    The ADCS requirements of various payloads on FS3 are stringent enough to demand some type of data
processing in order to meet attitude determination requirements. This paper detailed Kalman filtering, and more
specifically, the 7-state and 6-state Kalman filters. Both filters proved to meet the ADCS attitude determination
requirements successfully with little difference in achieved accuracy. However, the 6-state filter places much less
computational demand upon the on-board computer. Based upon this data, the 6-state filter is a more logical choice
for the FS3 ADCS filter. However, there is still much analysis that needs to be completed before a final verdict
might be reached.




                                                                                                                 15
                                                                                             American Institute of Aeronautics and Astronautics
                                                                 Appendix A.                                 Expanded 7-State Results
                                                                                               F alco n S at3 : A D C S L o g F ile
                                                                                                  7 S ta te A n g u la r E s tim a tio n E rro r


                       R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                   P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )


    3



    2



    1



    0



   -1



   -2



   -3




             19 T hu                     3A M                                   6A M                              9A M                         12P M                           3P M                                   6P M         9P M
A ug 2004                                                                                                                T im e



                                     Figure 13.                             7-State Angular Estimation Error, Expanded View
                                                                                                F alco n S at3 : A D C S L o g F ile
                                                                                              7 S ta te A n g u la r E s tim a tio n E rro r (D e ta il)


                          R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                  P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )
     0.4


     0.3


     0.2


     0.1


     0.0


    -0.1


    -0.2


    -0.3



            19 T hu                    3A M                                    6A M                               9A M                         12P M                            3P M                                   6P M         9P M
A ug 2004                                                                                                                   T im e



                         Figure 14.                             7-State Angular Estimation Error (Detail), Expanded View




                                                                  16
                                              American Institute of Aeronautics and Astronautics
                                                                               F alco n S at3 : A D C S L o g F ile
                                                                                   7 S ta te R a te s E s tim a tio n E rro r


                                W x E s t im a t io n E r r o r                                                                         W y E s t im a t io n E r r o r

     0.06


     0.05


     0.04


     0.03


     0.02


     0.01


     0.00


    -0.01


    -0.02


    -0.03

                 19 T hu          3A M                                  6A M                      9A M                          12P M      3P M                               6P M     9P M
A ug 2004                                                                                                 T im e



                              Figure 15.                          7-State Rates Estimation Error, Expanded View
                                                                               F alco n S at3 : A D C S L o g F ile
                                                                               7 S ta te R a te s E s tim a tio n E rro r (D e ta il)


                                     W x E s t im a t io n E r r o r                                                                        W y E s t im a t io n E r r o r



     0.00075


     0.00050


     0.00025


     0.00000


    -0.00025


    -0.00050


    -0.00075




               19 T hu          3A M                                   6A M                       9A M                          12P M        3P M                               6P M     9P M
A ug 2004                                                                                                     T im e



                           Figure 16.                 7-State Rates Estimation Error (Detail), Expanded View




                                                 17
                                 American Institute of Aeronautics and Astronautics
                                                                         Appendix B. Expanded 6-State Results
                                                                                                 F alco n S at3 : A D C S L o g F ile
                                                                                                     6 S ta te A n g u la r E s tim a tio n E rro r


                       R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                    P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )
    3



    2



    1



    0



   -1



   -2



   -3



                3P M         6P M                                    9P M                                19 T hu                         3A M                       6A M                                   9A M                   12P M
18 W ed A ug 2004                                                                                                           T im e



                                    Figure 17.                            6-State Angular Estimation Error, Expanded View
                                                                                                 F alco n S at3 : A D C S L o g F ile
                                                                                               6 S ta te A n g u la r E s tim a tio n E rro r (D e ta il)


                           R o ll A n g le E s t im a t io n E r r o r ( N o r m a liz e d )                                                                P it c h A n g le E s t im a t io n E r r o r ( N o r m a liz e d )

     0.25

     0.20

     0.15

     0.10

     0.05

     0.00

    -0.05

    -0.10

    -0.15

    -0.20

    -0.25

             3P M       6P M                                    9P M                                   19 T hu                         3A M                       6A M                                   9A M                     12P M
18 W ed A ug 2004                                                                                                             T im e



                        Figure 18.                             6-State Angular Estimation Error (Detail), Expanded View




                                                             18
                                             American Institute of Aeronautics and Astronautics
                                                                           F alco n S at3 : A D C S L o g F ile
                                                                                6 S ta te R a te s E s tim a tio n E rro r


                                    W x E s t im a t io n E r r o r                                                                        W y E s t im a t io n E r r o r

    0.06

    0.05

    0.04

    0.03

    0.02

    0.01

    0.00

    -0.01

    -0.02

    -0.03



                    3P M     6P M                               9P M                  19 T hu                        3A M           6A M                               9A M    12P M
18 W ed A ug 2004                                                                                       T im e



                                  Figure 19.                          6-State Rates Estimation Error, Expanded View
                                                                           F alco n S at3 : A D C S L o g F ile
                                                                           6 S ta te R a te s E s tim a tio n E rro r (D e ta il)


                                        W x E s t im a t io n E r r o r                                                                     W y E s t im a t io n E r r o r



     0.00075


     0.00050


     0.00025


     0.00000


    -0.00025


    -0.00050


    -0.00075




                3P M       6P M                              9P M                   19 T hu                         3A M            6A M                                9A M   12P M
18 W ed A ug 2004                                                                                         T im e



                           Figure 20.                     6-State Rates Estimation Error (Detail), Expanded View




                                                      19
                                      American Institute of Aeronautics and Astronautics

				
DOCUMENT INFO